Execution Plans

In AlertD, every time you ask a question, you’re not just triggering a simple lookup—you’re initiating a custom-built execution plan designed specifically for your prompt. These plans are generated and interpreted by agents, run by a purpose-built engine (not an LLM), and fully visible in the UI so you can follow what happened, step by step.

What’s a Plan?

A plan is a set of actions the agent creates to figure out the answer to your question. Think of it like a blueprint that defines what data to fetch, how to process it, and how to show it back to you.

Each plan is:

• Written by the agent selected for your question
• Executed by our internal engine (not an LLM)
• Interpreted by the same agent to give you a clear, human-readable result

You don’t need to build a plan manually—the agent handles that for you. But you can see and trace every step it took.

Plan Structure

Each plan consists of discrete steps, often visualized in a card-like structure. A typical plan might include:

1. Service Call

e.g., “List all S3 buckets across all regions”

• Purpose: Fetch raw data from your environment
• Source: Data can come from AlertD’s ingested data store for fast local queries, or from live AWS API calls when real-time data is needed
• Output: JSON data or metric results

2. Data Transformation

e.g., “Collect all results into a single array”

• Purpose: Process, filter, or aggregate data
• Type: Data manipulation step
• Output: Transformed dataset

3. Table or Visual Output

e.g., “Transform the array into a readable table”

• Purpose: Format data for human consumption
• Type: Presentation layer
• Output: Structured table or chart

4. LLM Pass (Optional)

e.g., “Analyze the results for insights”

• Purpose: Generate natural language interpretation
• Type: AI analysis
• Output: Observations and recommendations

Each step is clearly labeled and includes a status check (pass/fail), so you can see what worked and where things might’ve gone wrong.

Key Takeaways

• Plans are transparent – Every step is visible and inspectable
• Plans are custom – Built specifically for your question
• Plans are reliable – Executed by deterministic engine, not LLM
• Plans are educational – Learn how AlertD answers questions
• Plans are debuggable – Inspect data streams and errors

Understanding execution plans helps you:

• Trust AlertD’s answers
• Ask better questions
• Debug unexpected results
• Appreciate the complexity AlertD handles automatically

Next Steps

• Understanding Results – How to read the tables and insights AlertD generates
• Team Collaboration – Share plans and results across your team